Transformer-Based Multiomics Study Identifies Important Role of Glycine, Serine, and Threonine Metabolism Pathway in...

Rheumatoid arthritis (RA) is frequently complicated by anemia, which exacerbates disease progression and worsens clinical outcomes. This study aimed to characterize the molecular alterations associated with RA complicated by anemia (RA_ane) using a transformer-based machine learning architecture combined with multiomics aggregation. Retrospective clinical analysis demonstrated a high incidence of anemia in patients with RA, which strongly correlated with systemic inflammation (P < 0.001). Specifically, hemoglobin levels showed significant negative correlations with C-reactive protein and erythrocyte sedimentation rate. To investigate the underlying molecular landscape, we established a multiomics cohort comprising 113 healthy controls, 257 patients with RA, 88 patients with systemic lupus erythematosus, and 37 patients with gout. By aggregating untargeted metabolomic, microbiomic, and transcriptomic profiles and applying an algorithmic consensus across multiple machine learning models, we identified a reliable panel of RA_ane biomarkers. These included depleted metabolites (e.g., l-tryptophan and glyceric acid) and up-regulated genes (e.g., ALAS2, CA1, and HBB). Functional enrichment analysis identified the dysregulation of the glycine, serine, and threonine metabolism (GSTM) pathway as a central molecular signature of RA_ane. Furthermore, comparative analyses indicated that these GSTM alterations were specific to RA_ane, distinguishing it from anemia in systemic lupus erythematosus and gout. The core metabolic changes and the up-regulation of ALAS2 were successfully validated in an independent external cohort and by quantitative polymerase chain reaction. In addition, our diagnostic classifiers demonstrated strong predictive performance (area under the receiver operating characteristic curve up to 0.854). This study maps a multidimensional molecular network centered on the GSTM pathway, providing proof-of-concept insights that may inform future diagnostic approaches and mechanism-driven research for RA-associated anemia.